Is a single model enough? The systematic comparison of computational approaches for detecting populist radical right content

Germanophone test datasets and how their performance is affected by different modes of
text preprocessing. In addition to individual models, we examine the performance of 330
ensemble models combining the above-mentioned approaches for the dataset with a particularly high volume of noise. Our findings demonstrate that the DL models, in combination with more computationally intense forms of preprocessing, show the best performance among the individual models, but it remains suboptimal in the case of more noisy datasets. While the use of ensemble models shows some improvement for specific modes of preprocessing, overall, it mostly remains on par with individual DL models, thus stressing the challenging nature of computational detection of PRR content.

automated content analysis; populist radical right; supervised machine learning; neural networks; dictionaries; ensemble modeling

Is a single model enough? The systematic comparison of computational approaches for detecting populist radical right content Articles